University Projects
An AI Study Guide and an Assistive Dispenser
2023 – 2024 · Product owner · Prototyping lead
Two team projects at TUM: a study guide that only quizzes you on your own lecture notes, and a water dispenser you can use without your hands.
2024 · Product owner
StudyBuddy
Turns lecture PDFs into practice quizzes, with every answer grounded in the source material.
2023 · Prototyping lead
Automatic Water Dispenser
Detects the height of any cup, fills it hands-free and tracks daily intake for one person.
PROBLEM & USERS
StudyBuddy: generic AI tools invent content that is not in the lecture, so students cannot trust them for exam preparation. Water dispenser: people with limited dexterity, older people and children can struggle with manual dispensers, and most people underestimate how little they drink.
Who it’s for: University students preparing for exams, and people with limited dexterity or busy home routines.
CONSTRAINTS
- StudyBuddy: answers must stay within the uploaded lecture material
- StudyBuddy: lecture PDFs are messy, with tables and multi-column slides
- Dispenser: works with any cup and needs no fine motor control
- Dispenser: safe for children to use unsupervised
KEY DECISIONS
StudyBuddy: grounded quizzes
As product owner, I scoped the product around one promise: every quiz answer comes from the lecture. The team built a retrieval pipeline that finds the relevant passages before writing each question, plus PDF parsing for tables and multi-column slides.
Water dispenser: hands-free and aware
As prototyping lead, I built cup-height detection so the dispenser fills any cup without buttons, unlike dispensers that need specific cups or preset amounts. It also measures the dispensed volume to keep a daily total for one person, with reminders and an automatic shutoff.
WHAT SHIPPED
A web app that turns uploaded lecture PDFs into quizzes with knowledge-gap analytics, and a working dispenser prototype with cup detection, intake tracking and reminders.
At a glance
For
University students preparing for exams, and people with limited dexterity or busy home routines.
Key decision
Ground every quiz answer in the lecture; take the buttons out of dispensing
Result
Tools people can rely on without extra effort
Stack
- React
- Python
- RAG pipeline
- LLM APIs
- Embedded systems
- Sensors
RESULTS
- StudyBuddy: quiz content stays grounded in the source material, cutting hallucinations
- Dispenser: people with limited dexterity can get water on their own, and home users can see their daily intake
LEARNINGS
- Parsing decides retrieval: messy PDFs and chunk size changed answer quality the most
- Calibration matters: sensors have to cover different cup shapes and fill speeds
- Timing matters: reminders work better tied to daily routines than to fixed hourly alerts
NEXT STEPS
- StudyBuddy: adapt quiz difficulty to how a student performs
- Dispenser: profiles for more than one person